Application of an Improved Adaptive Unscented Kalman Filter in Vehicle Driving State Parameter Estimation

Abstract

In the research of estimating vehicle driving state parameters, the combination measurement unit of Global Navigation Satellite System and Inertial Navigation System (GNSS/INS) often faces challenges such as inaccurate models and reduced accuracy and robustness due to noise effects. To address these issues, this article applied characteristics of the state equation's innovation vector following a Gaussian distribution and the detection function following a Chi-square distribution. First, the properties of normal distribution are utilized to adaptively set the threshold value of the detection function to identify outliers in the measurement data. Subsequently, a novel adaptive unscented Kalman filter with Chi-square test (CAUKF) is designed, based on the adaptive window weight allocation and $Z$-score normalization, to correct abnormal data that do not conform to the characteristics of the innovation vector. Finally, comparative experiments on various algorithms are conducted using real-world data in terms of accuracy and robustness, and the results are analyzed in practical vehicle applications. The experimental results demonstrate that, without introducing noise errors in the target system, CAUKF exhibits superior accuracy compared to other algorithms. Moreover, in the testing of data contaminated with noise, CAUKF shows sensitivity to outlier data while ensuring rapid recovery of abnormal data without affecting data characteristics or calculating measurement noise characteristics. In summary, the CAUKF method effectively enhances the accuracy and robustness of the system.